Show HN: TUI-use: Let AI agents control interactive terminal programs
agents cursor
| Source: HN | Original article
A new open‑source project called **TUI‑use** landed on Hacker News on Monday, promising to let large‑language‑model agents drive interactive terminal programs the way a human would. The toolkit captures screen buffers, parses cursor positions and injects keystrokes, giving agents direct access to text‑based user interfaces (TUIs) such as Vim, Git’s interactive rebase, MySQL shells, and system monitors. Its core is a Go library that hooks into the pseudo‑terminal (PTY) layer, exposing a simple API that any LLM‑backed agent can call to “see” and “type” inside a live console.
The capability matters because most AI‑driven automation so far has been limited to one‑shot shell commands or API calls. Real‑world workflows often involve prompts, menus and live feedback that only a TUI can provide. By bridging that gap, TUI‑use enables agents to perform complex, stateful tasks—e.g., resolving merge conflicts, tuning performance parameters in ncurses dashboards, or guiding a user through a multi‑step installation—without human intervention. As we reported on April 9, Claude‑Managed Agents already demonstrated autonomous planning and execution; TUI‑use adds the missing “hands‑on” layer that turns planning into concrete interaction.
The next few weeks will reveal whether developers adopt the library for production agents. Key signals to watch are integrations with existing agent frameworks such as Claude‑Managed Agents, AutoBe’s code‑generation pipelines, and Monocle’s self‑healing loops. Security auditors will also scrutinise how the tool handles credential exposure and sandboxing, given its ability to drive privileged consoles. If the community can tame those risks, TUI‑use could become the de‑facto bridge that lets AI agents manage the full spectrum of command‑line tooling, reshaping DevOps, data‑science and remote‑work workflows across the Nordic tech scene.
Sources
Back to AIPULSEN